Improvement of non-negative matrix factorization-based active sonar reverberation suppression method using L1-norm and majorization-minimization

In this paper, we conduct research on an algorithm to suppress reverberation using a non-negative matrix factorization technique for target detection using an active sonar system. In this paper, we focus on the disadvantage that conventional algorithm estimates the target echo and reverberation bases through iterative estimation, which takes a long time to calculate. To improve this, we use the L1-norm to quickly converge to the desired solution at the beginning of the iteration. In addition, considering the fact that the heuristic multiplicative update method used in the conventional algorithm does not mathematically guarantee the convergence of additional penalty functions, the update equation is developed through a majorization-minimization technique that can guarantee convergence. Simulations were performed to verify the algorithm, and the results showed that the signal-to-noise ratio performance of the proposed algorithm was improved by 3 dB compared to the conventional algorithm.